3

I am working on a critically endangered bustard species in India (Houbaropsis bengalensis). I need to prepare a land cover map of its 'suitable' grassland habitat across its distributional range. I have points (and can construct polygons from them using Google Earth's help) where 'suitable' (for the species) grasslands exist since I did presence/not detected surveys in its distribution range.

Since the very same region also has a great amount of cultivation activity, sometimes land cover maps resulting just from the spectral signature using supervised classification tends to pick up croplands as 'grasslands'.

In order to avoid this obviously erroneous assessment, we (myself and collaborators) thought that using a temporal-spectral signature might be a better thing to do since croplands experience this sudden dip in NDVI when harvested and till new crops are sown (they may also have multiple cropping within a year), whereas grasslands do not (well...grasslands too are burnt once a year as a habitat management practice).

May I request people on this forum to help me in making a more detailed land cover map of this study area(s)/distribution range using temporal-spectral signatures in a supervised/unsupervised classification framework in order to distinguish between otherwise similar land cover types such as that in my case (natural/semi-natural grassland and cropland)? Is it even possible to do this in QGIS entirely, or perhaps using the interface of R and QGIS?

Please note that I would be using PROBA-V NDVI (333metre resolution) data for this.

  • I'm wondering why you would not use Landsat 8 OLI for this? LS 8 has the spectral resolution you would need to differentiate between crops. Also, you will need to use spectral data for your classification rather than NDVI alone. You can always incorporate NDVI into the classification as another band. – Aaron Mar 31 '16 at 1:06
  • @Aaron one guess as to why PROBA-V is the coverage. Swath-width and temporal revisit will allow mapping of larger areas than what is possible with OLI. – Mikkel Lydholm Rasmussen Mar 31 '16 at 18:38
  • Hello Aaron, I guess your concern has been well answered by Mikkel. One saves a lot of time 'data processing' time with PROBA-V data vis-a-vis OLI. – Rohit Jha Apr 1 '16 at 14:22
1

One way to take into account the plant phenology is to stack images from different parts of the growth season into one big file, and perform the (un)supervised classification on that. This will let you classify the area as a function of the plant growth across the entire season, rather than just one point in time.

The stacking of files can be done using a number of different tools, such as 'build virtual raster' or 'merge' in QGIS / GDAL.

  • Hi Mikkel, thank you for your response. I plan to do exactly that, stacking the images to take advantage of the difference in the way croplands are managed from natural/semi-natural grasslands. A very neat way of doing the same is using PROBA-V decant2 NDVI images of the region of interest for all months of the year. Can you expand on your answer about how exactly can this be done in QGIS? Is there an equivalent of ArcGIS's 'Iso Cluster' tool for unsupervised classification in QGIS? Please let me know. – Rohit Jha Mar 31 '16 at 16:39
  • @RohitJha try using the Semi-Automatic Classification Plugin in QGIS. There is plenty of guides and tutorials on the web to help you through the use. – Mikkel Lydholm Rasmussen Mar 31 '16 at 18:35
  • @RohitJha , do you have any results of your work? Would be interested in seeing how you solved the issue. Try to do exactly the same! – xmisx May 23 '17 at 13:11

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.